-
Notifications
You must be signed in to change notification settings - Fork 3.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add roadmap #67
Add roadmap #67
Conversation
geekan
commented
Mar 20, 2024
- add roadmap and goals.
- fix typo of the " evaluation" dir name.
Thank you! But the second point has been resolved. |
I'd suggest using checkboxes, so we can track our progress over time. Could probably also just use |
3. Confidential information mosaic: supports inquiry and mosaic of confidential information (such as passwords) | ||
3. Effect: Exceeding devin’s score on SWE-bench | ||
1. The trial-and-error process, that is, the agent generates both patch and test-patch (for self-test), and then the agent debugs it until it succeeds, and then generates the real patch, which may improve the effect by ~2 times | ||
4. Basic module |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Which programming language is selected ?
Yeah for the readme I'll make a simplest one. How about just using Project in GitHub? It can have things like checkboxes |
2. Memory | ||
1. Short-term memory: supports the storage and complete reading of all historical information within a time window | ||
2. Long-term memory: supports storage, retrieval, and weighted reordering of all historical information within a time window | ||
3. Program memory: supports symbolic storage and retrieval and weighted reordering of all historical information within a time window |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you describe the difference between "program memory" and "long-term memory"?
1. Note that it is recommended to pre-install a series of dependencies, such as chromium driver, to reduce the cost of repeated pulls. | ||
2. Support baremetal: Same as above | ||
3. Support vm: Same as above | ||
2. Memory |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is memory managed by individual agents? Or does opendevin keep its own memory? Maybe both?
1. Selenium: Natural interactions can occur, such as link clicks and jumps, page rendering, etc. | ||
2. Playwright: Same as above |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why support both?
1. SERPAPI: for fast structured interaction with different search engines | ||
2. SERPER: Same as above | ||
3. Duckduckgo: Fast structured interaction with ddg | ||
4. Google official API: fast structured interaction with google |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why not just use the regular browsing module for search?
1. Support GPT-4 | ||
2. Support GPT-4V | ||
3. Support ollama | ||
4. Support other APIs compatible with OpenAI |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
toggling between models should be pretty trivial, especially any OpenAI models, as well as claude (which supports a similar interface)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
toggling between models should be pretty trivial, especially any OpenAI models, as well as claude (which supports a similar interface)
By the way, can we add Gemini Pro from Google? It's free to use, and from a large, trusted provider.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We've got LiteLLM as our LLM-driver now, which should handle this
2. Use Case and interaction | ||
1. Document learning: Supports specifying a URL or keyword to learn documents, use corresponding documents, and use tools. | ||
2. Repair based on issue: Given a github repo and issue, the Agent can issue a Pull Request for Review | ||
3. Confidential information mosaic: supports inquiry and mosaic of confidential information (such as passwords) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I wonder if we can integrate with 1password or other password managers, rather than giving credentials directly to OpenDevin.
This seems like maybe a stretch feature--probably not one we want to hack in, and not critical for short-term goals
1. Community operation | ||
1. Establish an open source committee and clarify the board members and management mechanism. This committee is mainly used for SOP design and operation. | ||
2. Set up maintainer(s) and join some developers into the github group. They can determine which code can be incorporated and push everything forward. | ||
3. Design and provide community operation SOP documents, determine entry and exit criteria, and determine the responsibilities of different roles. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Really happy to see governance on the roadmap! 😄
This PR seems to have gotten a bit old. Should we consider closing it or revise it to match the current advance of the library? |
Agreed. Let's take another pass at it |